Welcome to Scratch-GAN
, a comprehensive repository dedicated to implementing various Generative Adversarial Networks (GANs) from scratch. This repository aims to serve as an educational resource for beginners and enthusiasts who are eager to learn about GANs and their diverse applications in the field of machine learning and artificial intelligence.
scratch-GAN
is designed to guide you through the fascinating world of GANs, starting from the very basics and gradually moving to more complex models. Each GAN model in this repository is implemented with detailed comments and explanations, making it easier for beginners to understand the core concepts and mechanics behind these powerful neural networks.
-
Basic GAN
: Implementation of the original GAN architecture for generating simple datasets like MNIST. -
DCGAN (Deep Convolutional GAN)
: Extending GANs with convolutional networks to improve image quality and stability. -
CGAN (Conditional GAN)
: Implementing conditional generative models for controlled image generation. -
InfoGAN
: Unsupervised learning with information-theoretic extensions to GANs. -
WGAN (Wasserstein GAN)
: Introducing Wasserstein loss for improved training stability and model evaluation. -
CycleGAN
: Implementing CycleGAN for unpaired image-to-image translation tasks. -
StyleGAN
andStyleGAN2
: Advanced GANs for generating high-quality, realistic images with style control. -
BigGAN
: Creating high-resolution images using large-scale GANs.